Eve library

Eve interactive artifacts — delivery and verification specification

Spec

Revision 5

On this page

Status and ownership#

Prepared 2026-10-09. Canonical execution plan for eve-interactive-artifacts, with render/rollout operations in platform-ops. This specification is submitted for independent review alongside the PRD and ADR-0083 successor. It describes required work, not a completed implementation. GOV.01 records approved revisions and binds final budgets; implementation remains dependency-gated until then.

Approved scope baseline (2026-10-09): Elikem explicitly approved the PRD and architecture in his signed-in Eve Workspace session using the Jerry Chrome profile. PRD document f23b42e4-6241-4fd7-af54-f4231fb26217 revision 3, content hash be94ccb9455d2ebc20e77bd0a8796a1177c771520e9ccc28d560622c8fa39add, approved at 2026-10-09T17:36:03.573Z. ADR successor a096a047-b6f8-4037-800e-6744517abffd revision 4, content hash 96732f00069cdc1429c9cecd661c9b1afce8460dc50ca8a043b009c04ed889d8, approved at 2026-10-09T17:36:19.611Z. Both approval records identify Elikem (666d8bfd-4497-4fd0-87c6-a391c5aa8440); content hashes are unchanged from the submitted revisions. This delivery specification remains subject to independent review. Its numerical budgets are proposed acceptance thresholds, not measured performance or an implemented-feature claim.

Product ownership: Eve interaction, assistant, content/design and data workers. Operational ownership: platform-ops and release controller. Independent code/document reviewers own acceptance review; Elikem retains governed document approval, publication, UAT, secrets administration and spending authority. Do not change automatic worker/reviewer/delegate switches or seat focus as part of this program.

Requirement coverage#

R1 adaptive answers: UX.01, UI.01-UI.07, M1, QA.03. R2 inspectable results: DATA.01, DATA.02, UI.03, ART.01, QA.01. R3 live dashboards: DATA.03-DATA.05, M2. R4 artifact lifecycle: ART.01-ART.03, ART.05, M1. R5 scenarios: UI.04, SIM.01, M4. R6 motion: MOTION.01-MOTION.05, M3. R7 templates/design systems: ART.04, selected DS prerequisites, QA.02. R8 collaboration: ART.05, DS07.1 and DS07.4, M4. R9 inputs/connectors: DATA.06-DATA.07, selected R15 tasks, M4. R10 exports: ART.06, MOTION.04, DS08.2 and DS08.3, M3. R11 surfaces/accessibility: UX.01, UI.06-UI.07, QA.02. R12 reliability/operations: ART.03, DATA.03, MOTION.04, QA.01, both platform tasks, FINAL. DOC.01 supplies accurate reviewed user/developer/operations documentation across all requirements.

Evidence and proposed budgets#

These are proposed acceptance budgets to ratify in GOV.01, not measured results. If measurements require a change, obtain a reviewed revision before changing the gate; never weaken a test to hide a failure.

Hard gates: every deterministic source/calculation/permission case passes; every displayed source-derived value has valid provenance or an explicit unavailable state; no unauthorized side effects or data disclosure; no false render/query success; no unresolved critical/high findings. Every unsupported client receives a safe useful fallback.

Initial bounds: 128 UI nodes with depth at most 8; charts at most 8 series and 1,000 displayed points per series; tables page at most 200 displayed rows; dependency views at most 200 nodes/500 edges with explicit truncation; file import at most 5 MiB and 10,000 rows until measured expansion; motion 15-60 seconds, at most 1080p/30 fps for the initial exporter. Source-side query row/time caps are independent of displayed limits. Validate the same caps server-side and client-side.

Interaction budget: local filter/slider/selection p95 at most 100 ms on the recorded reference device and dataset; usable component p95 at most 250 ms after receipt of a valid complete UI event; restore must not generate another model request. Query budget: approved representative Eve reads p95 at most 2 seconds, hard timeout at most 10 seconds. Render jobs must have measured resource limits, explicit progress and a cancel/cleanup contract; MOTION.01 sets the numerical render-time/memory admission after measuring the chosen toolchain. Test load must never compete unboundedly with the live board.

Model evaluation: at least 30 representative fixed prompts covering text-appropriate questions, visual exploration and controls/scenarios, plus deterministic malformed/adversarial cases. Predeclare scoring and expected sources. Target at least 90% first-pass valid admitted outputs and 100% safe fallback on the negative contract corpus; useful completion must be no worse than the text baseline, with at least 90% success on the agreed task set. Report actual counts and uncertainty; the small set is not a universal parity claim. Format-choice and design scoring require independent assessment. Report time-to-first-usable-answer separately from model wait, tool time, stream assembly and renderer time; report cost/tokens and bounded repair rate. Existing repo test-provider policy applies.

Reuse boundary#

8.5 owns long-turn controls; 8.14 rich reply rendering; 8.15 the base artifact workspace; 8.19 background operation delivery. Do not duplicate these features to satisfy the new family. Their consumers prove the specific reused path works at the candidate revision.

DS01.1/DS01.2/DS01.5/DS01.8 own the design kinds/schema, scene schema, revision patches and scoped BFF persistence. ART.01 extends these with dashboard/scenario/motion subtypes. The current design-only key boundary is retained; DS01.7's owner-only credential act stays with Elikem. DS07.1/DS07.4 own comments and concurrent human-agent edits. DS08.2/DS08.3 own export/handoff. DS12.1 owns design-system pins/defaults/locks; DS13.1 owns templates. R15.1/R15.3/R15.6 own connector registry, read/write admission and the first OpenRouter usage connector. Existing dependency declarations must be verified, including body-only missing edges; GOV.02 owns that reconciliation and activation.

DS11.3 supplies design-studio data binding when needed; the new DATA layer adds query provenance/refresh and must use an explicit adapter rather than a second source model. General code execution remains owned by 17.25 and is not a prerequisite for typed scenarios. Do not revive rejected public links, Stripe, broad frontier nodes or unrelated backlog work.

Required verification per implementation task#

Read actual implementation and its callers. Add focused contract/calculation/store/API tests when those layers change, with real domain expectations. Test doubles belong at dependency boundaries. Run touched-project typecheck and targeted tests; run actual Chromium rendering and changed-flow coverage for UI, with matching route contracts for new pages. Run native checks where the task names native behavior. Apply the heavy-verification skill and the remote-development policy before expensive checks. Code, servers, dependencies and build/test caches for Mac-origin development run in /root/oshun-mac-dev on 31.220.93.15, reached over SSH loopback forwards.

Evidence records the exact commit, commands/results, artifact/source revisions, real output files and material limitations. Independent review remains independent. Follow registered eval-data governance when touching registered cases or results. Commit and push code per repository policy; canonical product documents remain in Eve. Do not mark this program done from a status count, a schema, mocks alone or the presence of files.

Milestone order and operational exit#

M1: usable interactive answers plus persistence and permission enforcement across supported surfaces. M2: authoritative live delivery dashboard, refresh, source inspection and truthful failure states. M3: snapshot-to-editable-motion-to-decoded-export, preserving every source value. M4: real connector/imports, templates, calculations and human-agent collaboration. FINAL: all requirements reconciled with evidence, independent review accepted, tested main, normal Dev release, reviewed documentation and operational recovery. No UAT promotion is implied. Feature availability must reflect what is actually deployed and enabled.

If a prerequisite is missing, route the narrow blocker to its accountable task and continue independent work. Do not suppress a requirement or claim the feature is complete. GOV.01 and GOV.02 prevent code from racing unapproved decisions or unprepared prerequisites.

Task matrix#

eve-interactive-artifacts:GOV.01 — Ratify the interactive artifacts product scope and successor UI architecture#

Owner: Eve governance. Priority P1; effort S.

Obtain the independent decisions on the submitted PRD and successor to ADR-0083, reconcile requested changes, and record approved revision IDs and content hashes. The user authorized delivery planning on 2026-10-09; that is not self-approval of newly authored revisions. Keep all implementation tasks behind this gate. Preserve existing operator-only collaboration, admin/BFF placement, content authority and publication rules.

Acceptance:

  • Both governed documents have independently approved current revisions and any review changes are resolved.
  • The delivery specification names approved scope, proposed performance budgets, component/resource limits, connector admission, renderer licensing decision process and all milestone exits.
  • Evidence includes actual approval records and hashes; a draft or in-review revision cannot close this task.

Dependencies: none; governance entry point.

eve-interactive-artifacts:GOV.02 — Reconcile and activate the required existing artifact and design prerequisites#

Owner: Eve governance. Priority P1; effort S.

Read the current bodies, dependencies, decisions and evidence of the reused 8.5/8.14/8.15, design studio and connector tasks listed in the delivery specification. Reuse their work, add missing real prerequisite edges, and schedule only dependencies needed here. Resolve obsolete path assumptions as ordinary implementation planning; do not manufacture a new product engine. Route credentials or authority requirements to the existing owner task without widening privileges.

Acceptance:

  • Publish a reuse matrix naming each existing task, accepted output, actual next action and consumer in this family.
  • All reused prerequisite chains have real board edges, no cycles, no rejected task dependencies, and no hidden later-only prerequisite; each unavoidable external/human blocker has a named existing owner.
  • Do not claim historical shipped/imported status proves current acceptance; consumers must verify the relevant output. No blanket closures or unrelated backlog activation.

Dependencies: eve-interactive-artifacts:GOV.01.

eve-interactive-artifacts:UX.01 — Specify and review the working views and interaction behavior#

Owner: Eve experience. Priority P1; effort M.

Create a design brief and reviewable screen/interaction specifications for inline answers, saved dashboards, evidence inspection, controls, artifact history and motion editing. Use the frontend-skill app guidance and existing Oshun tokens. Cover admin first, customer-safe shell rendering and native/mobile sheets. Show the actual operator workflow without decorative dashboard mosaics.

Acceptance:

  • Visual thesis, content plan and interaction thesis are recorded in Eve with layouts for desktop and narrow/mobile screens.
  • Define loading, partial, empty, stale, denied, failed, retry, reconnect, cancellation and conflict states; identify primary action and keyboard/focus behavior.
  • Usefulness review includes a text-only alternative and explains when Eve should choose it; reduced-motion mode preserves the explanation.

Dependencies: eve-interactive-artifacts:GOV.01.

eve-interactive-artifacts:UI.01 — Implement the versioned interactive answer contract and validator#

Owner: Eve interactive answers. Priority P1; effort M.

Extend the confirmation-only generative UI contract with a new negotiated version for declarative metric, table, chart, comparison, timeline, diagram and parameter-control nodes. Retain v1 confirmation semantics. Define stable node IDs, typed bindings, interaction events, limits and text alternatives; accept no executable markup, arbitrary imports or caller-defined authority.

Acceptance:

  • Exact schema validation rejects unknown versions/components/properties, unsafe content, excessive nesting/size and invalid data bindings.
  • Old clients and v1 confirmations continue to work unchanged; invalid candidates produce a useful deterministic fallback.
  • Round-trip and hostile-payload specs exercise real validator behavior; touched-project typecheck passes.

Dependencies: eve-interactive-artifacts:GOV.02, eve-interactive-artifacts:UX.01.

eve-interactive-artifacts:UI.02 — Stream and restore interactive answers without replay corruption#

Owner: Eve interactive answers. Priority P1; effort M.

Extend the existing SSE turn protocol with complete validated node/spec operations, monotonic revisions, completion and failure semantics. Persist enough state for reload and reconnect. Build a small compiler/assembler over the allowed component registry; do not require the full response before rendering a valid component.

Acceptance:

  • Interrupted streams, duplicate/out-of-order events, obsolete revisions and replay cannot duplicate nodes or attach them to the wrong turn.
  • Reload and reconnect restore the same accepted UI state; cancellation stops further mutations and presents partial completion honestly.
  • Measure time from accepted component event to usable UI; record model waiting time separately.

Dependencies: eve-interactive-artifacts:UI.01, eve-sota-gap-closure:8.5.

eve-interactive-artifacts:UI.03 — Render accessible metrics, tables, charts, comparisons and diagrams#

Owner: Eve interactive answers. Priority P1; effort L.

Implement trusted renderers for the admitted visual nodes using shared tokens and semantics. Include chart legends/axes/units, sortable tables with bounded rows, timelines, selectable dependency diagrams and textual/data-table alternatives. Distinguish unavailable data, observed zero and an empty result.

Acceptance:

  • Real fixture values, units, dates and category labels survive rendering and interaction exactly.
  • Keyboard and assistive-technology semantics, zoom, light/dark themes, narrow layout and reduced motion pass focused checks.
  • Large graphs/tables are bounded and explicitly disclose truncation; actual Chromium flows and visual captures verify layout and behavior.

Dependencies: eve-interactive-artifacts:UI.01, eve-interactive-artifacts:UX.01, eve-sota-gap-closure:8.14.

eve-interactive-artifacts:UI.04 — Implement typed filters and controls with authoritative action binding#

Owner: Eve interactive answers. Priority P1; effort M.

Add typed selection, date-range, number, slider, form and drill-down interactions. Local view changes remain local; server queries and writes use named bindings and existing authorization/confirmation. Preserve control state in history and send a concise state summary on follow-up turns.

Acceptance:

  • Changes to filters, selections and forms update the correct component without a model call for every interaction.
  • Forged events, cross-session/action replay and expired permissions are rejected server-side; a display node cannot grant a mutation.
  • Input validation, focus retention, reset, share-free persistence and stale-control recovery are covered by integration/browser tests.

Dependencies: eve-interactive-artifacts:UI.01, eve-interactive-artifacts:UI.03.

eve-interactive-artifacts:UI.05 — Teach Eve to select, compose and revise useful interactive responses#

Owner: Eve interactive answers. Priority P1; effort L.

Add model tools/prompt policy to propose a validated response specification from authorized tool results. Choose text when it serves the question best; choose interactive output for comparison, exploration and useful controls. Bind values to result handles; preserve node identity across follow-up edits. Use existing model routing and bounded repair/fallback.

Acceptance:

  • Representative prompts produce useful text or the appropriate UI with sources and clear assumptions; no invented data or unsupported tools.
  • A malformed candidate gets a bounded repair or honest text fallback; timeouts and unavailable models do not become success.
  • Follow-up requests revise the intended view and keep source/permission bindings; token/cost and repair counts are measured on the required inexpensive OpenRouter model.

Dependencies: eve-interactive-artifacts:UI.02, eve-interactive-artifacts:UI.03, eve-interactive-artifacts:UI.04.

eve-interactive-artifacts:UI.06 — Integrate interactive answers into admin and customer web chat#

Owner: Eve interactive answers. Priority P1; effort L.

Wire the complete feature through real admin and customer web chat flows, capability negotiation, history, follow-up turns and evidence inspection. Operator-only data/tools remain available only through the authorized admin/BFF path. Rendering capability does not grant access to the operator board.

Acceptance:

  • Actual chat requests generate and update usable views on both supported web surfaces, with server permissions enforced.
  • History/reload, cancellation, fallback and one follow-up edit work in Chromium at desktop and narrow widths.
  • New page routes, if needed, receive matching route-matrix rows; analytics distinguish generation, rendering and interaction failures.

Dependencies: eve-interactive-artifacts:UI.05, eve-sota-gap-closure:8.14.

eve-interactive-artifacts:UI.07 — Deliver mobile interactive reply and artifact parity#

Owner: Eve interactive answers. Priority P1; effort L.

Implement the same negotiated semantics in mobile and admin-mobile where those surfaces expose the assistant. Use native/sheet presentations and accessible equivalents for dense charts/diagrams. Define capability fallback for unsupported clients rather than silently dropping controls.

Acceptance:

  • Targeted native tests and a real supported simulator/device walkthrough prove filtering, evidence inspection, history and follow-up edits.
  • Touch targets, screen-reader order, large text, keyboard avoidance and reduced motion remain usable.
  • Offline/stale results carry timestamps and cannot execute unauthorized stale actions; product scope determines which data each mobile surface can read.

Dependencies: eve-interactive-artifacts:UI.05, eve-sota-gap-closure:8.14.

eve-interactive-artifacts:ART.01 — Extend artifact revisions with immutable data and source bindings#

Owner: Eve artifact lifecycle. Priority P1; effort M.

Extend the existing artifact workspace with response/dashboard/scenario/motion types. Keep editable specification revisions distinct from immutable observation snapshots and render outputs. Record owner, origin turn, schema version, source IDs/revisions, query version, parameters, observedAt, content hashes, design pin and renderer version. Canonical saved operator artifacts are versioned design-format subtypes in Eve content through the existing DS01 scoped BFF path; sessions keep references. Do not expand a design-only key to PRD/ADR or publication authority.

Acceptance:

  • Refreshing data never rewrites the artifact specification or an already exported snapshot.
  • Optimistic revision checks, idempotency, lineage, migration/backward reading and failed-save recovery pass persistence integration tests.
  • Every source-bound value resolves to a valid authorized snapshot or explicit unavailable state; immutable revisions can be reproduced.

Dependencies: eve-interactive-artifacts:UI.01, eve-sota-gap-closure:8.15, eve-gap-analysis:DS01.1, eve-gap-analysis:DS01.2, eve-gap-analysis:DS01.5, eve-gap-analysis:DS01.8.

eve-interactive-artifacts:ART.02 — Save, reopen, compare and edit conversational artifacts#

Owner: Eve artifact lifecycle. Priority P1; effort M.

Connect inline answers to the artifact side panel and history. Save a view, reopen it in later conversations, compare revisions, fork a scenario, and target an Eve edit to a stable node. Preserve the existing explicit publish decision and owning-domain save behavior.

Acceptance:

  • An operator can ask, save, leave, reopen, edit and compare a dashboard without losing filters, sources or attribution.
  • A stale edit produces a visible conflict and recovery path; edits never silently overwrite another revision.
  • Browser coverage starts from the real chat UI and verifies stored results after reload.

Dependencies: eve-interactive-artifacts:ART.01, eve-interactive-artifacts:UI.06.

eve-interactive-artifacts:ART.03 — Enforce artifact permissions, revocation, retention and deletion#

Owner: Eve artifact lifecycle. Priority P1; effort M.

Carry authorization through saved artifacts, cached snapshots, queries, media, exports and job results. Preserve the one-human operator boundary and existing product isolation. Recheck permissions on access and refresh; apply retention/deletion to derived artifacts and stop future refresh/render jobs for revoked sources.

Acceptance:

  • Cross-user/product/session access, guessed IDs, stale credentials and shared cache keys cannot disclose data.
  • Revocation and deletion invalidate reads/downloads and queued work consistently; immutable published retention follows approved policy with explicit conflicts.
  • Security integration tests include unauthorized export, provenance detail, cached chart and post-revocation refresh cases.

Dependencies: eve-interactive-artifacts:ART.01.

eve-interactive-artifacts:ART.04 — Apply reusable dashboard and explainer templates with design-system pins#

Owner: Eve artifact lifecycle. Priority P2; effort M.

Extend the existing design-template and publication mechanism with dashboard, scenario and explainer starters. Bind the selected design-system version and expose approved layout choices. Reuse the token source of truth and studio contracts; do not create a second template store or mutate tokens from model output.

Acceptance:

  • Each starter creates a valid editable artifact with honest empty data bindings and accessible layout.
  • Published design pins remain stable across refresh and export; locked defaults reject unapproved alternatives.
  • Changing a shared template does not silently mutate saved artifact revisions; migration/fork choices are explicit.

Dependencies: eve-interactive-artifacts:ART.02, eve-gap-analysis:DS13.1, eve-gap-analysis:DS12.1.

eve-interactive-artifacts:ART.05 — Extend anchored comments and agent co-editing to interactive artifacts#

Owner: Eve artifact lifecycle. Priority P2; effort M.

Reuse existing comments and concurrent-edit semantics for dashboard nodes, calculation assumptions and motion scenes. Collaboration means Elikem and agent lanes. Preserve actor/origin attribution, presence expiry, disjoint-patch rebase and explicit conflict resolution; no public links or new human roles.

Acceptance:

  • Comments survive node movement; deleted targets become visibly orphaned and keep history.
  • Disjoint agent and human edits can merge; overlapping edits are refused with actionable conflict detail and no lost update.
  • Presence is bounded/ephemeral; replay, disconnect and crash-recovery browser/integration cases pass.

Dependencies: eve-interactive-artifacts:ART.02, eve-gap-analysis:DS07.1, eve-gap-analysis:DS07.4.

eve-interactive-artifacts:ART.06 — Export reproducible artifact snapshots and editable handoffs#

Owner: Eve artifact lifecycle. Priority P1; effort L.

Extend the existing export/handoff pipeline for interactive answers, dashboards and scenarios. Support a data export and portable specification bundle, plus PDF/PNG/SVG and editable deck handoff where the existing format supports it. Record unsupported features explicitly. Downloads freeze a source snapshot; public hosting is not introduced.

Acceptance:

  • Export files open and reproduce the pinned values, labels, fonts and layout within declared format limits.
  • Exports include provenance, observation time, design/version pins and a manifest; no credentials or hidden unauthorized rows.
  • Round-trip import or agent handoff reconstructs supported editable content; visual and semantic export comparisons use actual generated files.

Dependencies: eve-interactive-artifacts:ART.01, eve-interactive-artifacts:ART.03, eve-gap-analysis:DS08.2, eve-gap-analysis:DS08.3.

eve-interactive-artifacts:DATA.01 — Define the governed metric and query registry for Eve data#

Owner: Eve live dashboards. Priority P1; effort M.

Create versioned named query and metric definitions for task status, reviews, releases and dependency blockers. Specify filter types, permissions, date/time boundaries, units, status definitions, pagination, truncation and denominator semantics. Reuse authoritative board/content APIs and graph semantics; avoid broad analytics queries against the live board.

Acceptance:

  • Fixture cases distinguish shipped, verified, blocked, ready, waiting review, failure, missing history and actual zero.
  • Definitions are reviewable and inspectable; unsupported metrics are rejected instead of approximated silently.
  • Query inputs are allowlisted and parameterized, with bounded cost; historical claims require a real historical source.

Dependencies: eve-interactive-artifacts:GOV.02, eve-interactive-artifacts:UI.01.

eve-interactive-artifacts:DATA.02 — Execute dashboard queries with scoped access and inspectable provenance#

Owner: Eve live dashboards. Priority P1; effort L.

Implement a bounded read executor that returns values plus query identity/version, normalized parameters, observation time, source records/revisions and warnings. Use authorized APIs or dedicated read models; direct diagnostics against Eve PostgreSQL must be READ ONLY with statement timeout and named application identity, never a board lock.

Acceptance:

  • Known datasets produce exact aggregates and correct pagination/filter/time semantics; unauthorized sources and over-budget queries fail closed.
  • Selecting any number exposes its definition, source and refresh time, with authorized details and safe query representation.
  • Timeouts, partial sources and stale caches are explicit and cannot masquerade as a successful fresh zero.

Dependencies: eve-interactive-artifacts:DATA.01, eve-interactive-artifacts:ART.01, eve-interactive-artifacts:ART.03.

eve-interactive-artifacts:DATA.03 — Refresh saved dashboards reliably with bounded jobs and caches#

Owner: Eve live dashboards. Priority P1; effort L.

Add manual refresh and explicit opt-in refresh schedules using the existing job/loop runtime where applicable. Deduplicate runs, authorize each execution, coalesce identical reads safely, track job status and atomically replace the latest successful observation pointer. A failed refresh retains the previous snapshot with a stale/error label.

Acceptance:

  • Concurrent refresh, retry, restart, revoked credentials, deleted artifact and missed schedules do not duplicate side effects or leak results.
  • Visible per-chart last-success time and refresh error remain accurate; cache keys include permission scope and query/parameter versions.
  • Backoff, time/resource caps, cancellation and cleanup are tested with a controllable clock and real persistence.

Dependencies: eve-interactive-artifacts:DATA.02, eve-sota-gap-closure:8.19.

eve-interactive-artifacts:DATA.04 — Build dashboard authoring, filtering and evidence drill-down#

Owner: Eve live dashboards. Priority P1; effort L.

Allow Eve and the operator to assemble, resize/reorder within approved layouts, filter and inspect a saved dashboard. A follow-up such as compare last week or group by product creates a validated spec change or filter change with preserved provenance. Use admin studio/BFF placement and existing artifact views.

Acceptance:

  • The real UI creates a dashboard from a question, adjusts filters and layout, inspects a metric and saves/reopens it.
  • Contradictory filters, unavailable history, oversized queries and unsupported chart mappings show useful errors or alternatives.
  • The working view uses readable hierarchy and does not render a mosaic of ornamental cards; desktop/mobile behavior matches the brief.

Dependencies: eve-interactive-artifacts:DATA.02, eve-interactive-artifacts:ART.02, eve-interactive-artifacts:UI.04.

eve-interactive-artifacts:DATA.05 — Ship the delivery blockers dashboard using authoritative Eve records#

Owner: Eve live dashboards. Priority P1; effort M.

Deliver the first complete operator workflow: show what is blocking delivery across Oshun. Include scoped counts, task/review/release status, dependency graph, responsible lane and source links. Explain causal dependency blockers separately from correlated failures and distinguish no data from no blockers.

Acceptance:

  • A seeded integration environment has independently calculated expected counts and blocker paths; displayed and refreshed results match them.
  • A real authenticated Dev walkthrough saves the view, changes underlying data through normal product actions, refreshes and reopens the updated snapshot.
  • The feature never claims a task or release is complete merely from historical imported board status; all drill-downs work.

Dependencies: eve-interactive-artifacts:DATA.03, eve-interactive-artifacts:DATA.04.

eve-interactive-artifacts:DATA.06 — Import bounded CSV and JSON datasets with explicit schemas#

Owner: Eve live dashboards. Priority P2; effort M.

Add local CSV/JSON data sources for dashboards and what-if tools. Parse bounded inputs, infer candidate types for confirmation, preserve column names/units/nulls/time zones, and store data under the approved artifact storage/retention contract. Reuse DS11.3 bindings when connecting an artifact to the design studio.

Acceptance:

  • Malformed encodings, formula injection in exports, deeply nested JSON, duplicate columns, ambiguous dates and oversized files are handled explicitly.
  • Source hashes and schema choices are saved; reimport creates a new observation with a visible schema-drift decision.
  • A real file-to-chart-to-export workflow preserves values and obeys source deletion/access rules.

Dependencies: eve-interactive-artifacts:DATA.02, eve-interactive-artifacts:ART.03.

eve-interactive-artifacts:DATA.07 — Bind governed connectors to dashboards and prove the first live connector#

Owner: Eve live dashboards. Priority P2; effort L.

Add a dashboard data-source adapter over the existing operator connector registry, starting with the authorized read-only OpenRouter usage connector. Use documented actual fields, units, provider time windows and rate limits. Unsupported per-model or historical breakdowns must be unavailable, not estimated. Provide the extension contract for later warehouses/CRM without promising unbuilt adapters.

Acceptance:

  • One real credential-backed read populates a saved dashboard and can refresh through the governed connector, with redacted evidence.
  • Credential absence/rotation/revocation, provider errors/rate limits and schema drift are covered; tokens never enter model prompts, snapshots or exports.
  • The source connector has its required review record; no new paid account, connector activation or credential privilege is created without the existing owner authority.

Dependencies: eve-interactive-artifacts:DATA.03, eve-gap-analysis:R15.1, eve-gap-analysis:R15.3, eve-gap-analysis:R15.6.

eve-interactive-artifacts:SIM.01 — Provide reproducible what-if tools with explicit assumptions#

Owner: Eve scenario tools. Priority P1; effort M.

Implement a versioned deterministic calculation registry and parameter schema. First tools: a capacity/workload scenario using declared throughput assumptions and a resource-cost calculator using supplied unit costs. Distinguish observed inputs, user assumptions and scenario outputs; do not imply a forecast without a validated forecasting method.

Acceptance:

  • Arithmetic, unit conversion, invalid/missing inputs, zero/negative values, rounding and limits are tested against independent expected results.
  • Slider changes recompute locally or through bounded named calculations without model calls; p95 interaction target is measured.
  • Saving/forking/reopening reproduces outputs from the pinned formula version and inputs; updating source data requires an explicit scenario refresh.

Dependencies: eve-interactive-artifacts:UI.04, eve-interactive-artifacts:DATA.02, eve-interactive-artifacts:ART.02.

eve-interactive-artifacts:MOTION.01 — Select and prove the deterministic motion renderer and export toolchain#

Owner: Eve motion. Priority P1; effort M.

Run a bounded engineering spike comparing existing reusable presentation/design rendering with a parameterized React/Remotion option and a suitable deterministic alternative. Evaluate browser preview, editable source, MP4 encoding, captions, fonts, reproducibility, dependency security, licensing and self-hosted CPU/memory cost. Pin the chosen versions and record the decision in Eve; no purchase is authorized.

Acceptance:

  • Render a real 30-second chart/text/shape sample and inspect browser preview plus MP4 frames, duration and decoded output.
  • Document measured time/memory, fidelity limits, license applicability, version pins and a resource-bounded worker design.
  • Choice is reviewed before dependent implementation; if a paid license is required, identify its owner decision or choose a valid alternative.

Dependencies: eve-interactive-artifacts:GOV.02, eve-interactive-artifacts:UX.01.

eve-interactive-artifacts:MOTION.02 — Implement the editable motion scene and timeline contract#

Owner: Eve motion. Priority P1; effort M.

Add a versioned deterministic timeline of scenes containing text, shapes, charts and authorized image/media references. Define units, frame/time semantics, transitions, captions/transcript, source pins and reversible patch operations. Reuse existing presentation identities and design nodes through explicit adapters.

Acceptance:

  • Schema and timeline tests cover exact durations, frame boundaries, overlaps, invalid assets, text/number preservation and stable IDs.
  • Scene values bind to frozen snapshots and source document revisions; later data refresh cannot change an existing render.
  • No arbitrary model-authored code or remote asset URLs execute in the trusted renderer; migrations and fallback are defined.

Dependencies: eve-interactive-artifacts:MOTION.01, eve-interactive-artifacts:ART.01.

eve-interactive-artifacts:MOTION.03 — Build the motion preview editor and conversational revision flow#

Owner: Eve motion. Priority P1; effort L.

Implement play/pause/scrub, scene selection, timing changes, direct text edits and targeted conversational changes in the admin authoring surface. Display sources and editable captions/transcript. Keep preview output synchronized with a validated saved revision.

Acceptance:

  • A user can make and preview a 30-second explanation, change one number/word/timing, undo it and reopen the correct revision.
  • Preview and timeline agree frame-for-frame on tested scenes; missing fonts/assets and overflow are visible failures.
  • Keyboard access, reduced-motion storyboard and mobile playback/inspection work; the operator studio remains behind admin/BFF authorization.

Dependencies: eve-interactive-artifacts:MOTION.02, eve-interactive-artifacts:ART.02, eve-interactive-artifacts:UI.05.

platform-ops:eve-artifact-render-workers-2026-10-09 — Provision bounded private workers for Eve artifact rendering#

Owner: platform-ops. Priority P1; effort M.

Observed: the approved program needs durable CPU-bounded motion/export jobs; existing presentation-builder-runner.ts describes a local operator seam and is not evidence of a deployed motion service. Evidence: source inspection on 2026-10-09 and the MOTION.01 measured renderer decision. Impact: MOTION.04 cannot deliver reliable self-hosted exports until the worker runtime exists. Suspected cause: new product capability has no verified provisioned runtime. What would close it: provision the selected pinned toolchain under existing fleet/release ownership, private loopback/service networking, authenticated job and media access, bounded concurrency, scratch cleanup and resource supervision.

Acceptance:

  • Check fleet resources and active work first; deploy no public development ports and do not borrow another identity or grant scopes.
  • An authenticated test job produces a real file; resource exhaustion, restart, cancellation and unavailable dependencies fail observably.
  • Document service ownership, pinned image/tool versions, resource limits, health checks, cleanup and rollback; preserve existing Compose/remote-dev policy.

Dependencies: eve-interactive-artifacts:MOTION.01.

eve-interactive-artifacts:MOTION.04 — Render MP4 jobs with captions, cancellation and durable delivery#

Owner: Eve motion. Priority P1; effort L.

Implement bounded asynchronous render jobs on the selected self-hosted renderer. Render from immutable artifact/snapshot/media pins; store MP4, poster and captions with manifest hashes. Reuse existing operation status, cancellation and conversation result delivery. Stage publication separately from draft generation.

Acceptance:

  • Real encoded files decode with expected duration, dimensions, fps, captions and sampled visual/semantic fidelity.
  • Duplicate submission, cancellation, restart, missing assets, quota/time/memory limits and failed encoding never yield a false success or partial published download.
  • Jobs use restricted network and filesystem access, clean scratch outputs and expose measured resource/cost data where available.

Dependencies: eve-interactive-artifacts:MOTION.02, eve-interactive-artifacts:ART.03, eve-sota-gap-closure:8.19, platform-ops:eve-artifact-render-workers-2026-10-09.

eve-interactive-artifacts:MOTION.05 — Turn a saved dashboard or presentation into an editable explainer#

Owner: Eve motion. Priority P1; effort M.

Compose the complete workflow from a pinned dashboard snapshot or approved document/presentation revision to a short storyboard, editable motion preview and MP4 export. Keep claim/source mapping and generated narrative inspectable. Resolve number or source changes before render; offer explicit regeneration when the source advances.

Acceptance:

  • The delivery blockers dashboard becomes a 30-second explainer whose numbers and labels match the selected snapshot exactly.
  • A presentation/document example preserves source revision and factual claims through a targeted edit and export.
  • No synthetic footage or generative video model is needed for this feature; unavailable narration is explicit and optional, with a useful captioned silent export.

Dependencies: eve-interactive-artifacts:MOTION.03, eve-interactive-artifacts:MOTION.04, eve-interactive-artifacts:DATA.05.

eve-interactive-artifacts:QA.01 — Independently verify data correctness, isolation and hostile artifact behavior#

Owner: independent verification with owning workers. Priority P1; effort L.

Run an independent integration and adversarial assessment against the actual implementation and resolve findings through owning tasks. Cover query correctness, metrics, scenario math, provenance, artifact permissions, untrusted content, connector injection, export data leakage, replay, cancellation and sandbox escape attempts. Reuse registered evaluation governance when changing registered artifacts.

Acceptance:

  • All hard correctness/security cases pass on the candidate commit and pinned document revisions; failures retain reproducible evidence and owners.
  • Fixtures mock dependencies only, never the unit under test; live-source evidence complements rather than replaces isolated deterministic checks.
  • No bypasses, fabricated values, hidden fallback success or unresolved critical/high findings remain. Author and reviewer identities are distinct.

Dependencies: eve-interactive-artifacts:DATA.05, eve-interactive-artifacts:DATA.06, eve-interactive-artifacts:DATA.07, eve-interactive-artifacts:SIM.01, eve-interactive-artifacts:MOTION.05, eve-interactive-artifacts:ART.05, eve-interactive-artifacts:ART.06.

eve-interactive-artifacts:QA.02 — Verify cross-surface accessibility, visual quality and export fidelity#

Owner: independent verification with owning workers. Priority P1; effort L.

Exercise representative end-to-end flows on admin/customer web and supported native surfaces. Assess keyboard, screen reader, large text, contrast, reduced motion, responsive density, loading/error states and exported document/video fidelity. Keep customer surfaces isolated from operator data.

Acceptance:

  • Chromium journey tests and real native walkthroughs cover interaction, save/reopen, conflict, refresh and playback.
  • Visual captures and decoded export samples show readable content without clipping, wrong fonts or missing values.
  • Manual assistive-technology findings are recorded and fixed; automated scans alone do not claim full accessibility.

Dependencies: eve-interactive-artifacts:UI.07, eve-interactive-artifacts:ART.04, eve-interactive-artifacts:ART.05, eve-interactive-artifacts:ART.06, eve-interactive-artifacts:MOTION.05.

eve-interactive-artifacts:QA.03 — Measure usefulness, latency, reliability and model cost against text answers#

Owner: independent verification with owning workers. Priority P1; effort M.

Implement a bounded representative evaluation of response-format choice, correctness, completeness, design clarity and task completion against the current text-only baseline. Include simple questions where text should win, data exploration, comparisons, malformed output and unavailable sources. Use the stipulated inexpensive OpenRouter test route; stronger models require separate spend authorization.

Acceptance:

  • Predeclared rubric, independent expected outputs, fixed environment and model/version are recorded before results.
  • Meet approved thresholds for valid/fallback rendering, task usefulness, local interaction latency, time-to-first-usable-component, query latency and resource use; separate model wait from UI assembly.
  • Report tokens/cost, sample size, distributions, repair rate and limitations. No invented improvement percentage, frontier parity claim or cherry-picked demonstration closes the gate.

Dependencies: eve-interactive-artifacts:UI.06, eve-interactive-artifacts:UI.07, eve-interactive-artifacts:DATA.05, eve-interactive-artifacts:SIM.01, eve-interactive-artifacts:MOTION.05.

eve-interactive-artifacts:DOC.01 — Publish reviewed usage, implementation and recovery documentation#

Owner: Eve documentation. Priority P1; effort M.

Update canonical Eve documentation for asking, saving, refreshing, inspecting, editing, exporting and recovering artifacts. Explain supported connectors, capability availability, source freshness, scenario assumptions, motion limits and mobile differences. Document schemas, migration, troubleshooting and operational runbooks without restoring retired Markdown sources.

Acceptance:

  • Every advertised feature is linked to actual shipped behavior and a verified journey; missing providers/capabilities are described accurately.
  • Required document revisions are submitted and independently reviewed; release controller selection pins the approved revisions before code/tests that require them.
  • Relevant route matrices, API schemas and product capability inventories are updated in code commits as required.

Dependencies: eve-interactive-artifacts:DATA.07, eve-interactive-artifacts:MOTION.05, eve-interactive-artifacts:ART.04, eve-interactive-artifacts:ART.05, eve-interactive-artifacts:ART.06, eve-interactive-artifacts:QA.03.

eve-interactive-artifacts:M1 — Accept interactive answers and persistent artifact editing end to end#

Owner: Eve milestones. Priority P1; effort S.

Integration acceptance only: prove the new typed answers and artifact lifecycle together after the leaf tasks close. Review the actual implementation evidence, not only task statuses.

Acceptance:

  • Ask for an appropriate interactive response, operate controls, inspect sources, save, reload, edit and compare versions across supported surfaces.
  • Unsupported clients receive a useful fallback and permissions are enforced; any gap returns to its owning leaf.

Dependencies: eve-interactive-artifacts:UI.06, eve-interactive-artifacts:UI.07, eve-interactive-artifacts:ART.02, eve-interactive-artifacts:ART.03.

eve-interactive-artifacts:M2 — Accept the live delivery dashboard and source inspection workflow#

Owner: Eve milestones. Priority P1; effort S.

Integration acceptance only: prove the live blockers dashboard from question through refresh and evidence inspection.

Acceptance:

  • Change authoritative fixture records through product APIs, refresh the saved view and independently reconcile every affected result.
  • Demonstrate stale/error handling, no-data semantics and permission revocation without losing the prior truthful snapshot.

Dependencies: eve-interactive-artifacts:M1, eve-interactive-artifacts:DATA.05.

eve-interactive-artifacts:M3 — Accept the dashboard-to-motion and reproducible export workflow#

Owner: Eve milestones. Priority P1; effort S.

Integration acceptance only: prove saved evidence becomes editable motion and export without losing source lineage.

Acceptance:

  • Pin a snapshot, produce/edit a 30-second explainer and decode the exported MP4; numbers and captions match the chosen revision.
  • Cancellation, retry and data changes after snapshot creation behave as specified; export manifests permit reproduction.

Dependencies: eve-interactive-artifacts:M2, eve-interactive-artifacts:MOTION.05, eve-interactive-artifacts:ART.06.

eve-interactive-artifacts:M4 — Accept templates, scenarios, connectors and agent collaboration#

Owner: Eve milestones. Priority P1; effort S.

Integration acceptance only: prove all remaining recommended capabilities together and audit the reuse of existing design/connector work.

Acceptance:

  • Create from an approved template, import data, bind the live connector, fork a scenario and perform disjoint/conflicting human-agent edits.
  • No unsupported external sharing, unauthorized provider access or second template/connector authority was introduced.

Dependencies: eve-interactive-artifacts:M3, eve-interactive-artifacts:DATA.06, eve-interactive-artifacts:DATA.07, eve-interactive-artifacts:SIM.01, eve-interactive-artifacts:ART.04, eve-interactive-artifacts:ART.05.

platform-ops:eve-artifact-rollout-readiness-2026-10-09 — Verify Eve artifact monitoring, recovery and controlled Dev rollout#

Owner: platform-ops. Priority P1; effort M.

Observed: new refresh/render/export services need operational acceptance before activation. Evidence: the program requires real scheduled reads and render jobs, with no current production evidence claimed. Impact: an unbounded queue, stale source, failed job or incompatible migration could harm Eve availability or mislead users. Suspected cause: new capability requires explicit rollout and recovery coverage. What would close it: instrument and exercise the approved capability flags, queue/resource limits, freshness/error metrics, redacted logs, backup/restore and reversible migration/rollback on the candidate.

Acceptance:

  • Normal Dev release passes affected code/content gates at the recorded commit; flags reflect actual availability and do not bypass approval.
  • Exercise queue failure, worker restart, lost source access, retention cleanup and rollback; alerts and runbook actions are accurate with no fabricated counters.
  • No UAT promotion, public sharing, spending or privileged key change occurs without Elikem authority; record any such narrow owner action explicitly.

Dependencies: eve-interactive-artifacts:M4, eve-interactive-artifacts:QA.01, eve-interactive-artifacts:QA.02, eve-interactive-artifacts:QA.03, eve-interactive-artifacts:DOC.01.

eve-interactive-artifacts:FINAL — Close the program only after verified delivery and operational handoff#

Owner: Eve milestones. Priority P1; effort M.

Final integration and independent release acceptance only. Reconcile every requirement and task with implementation, test, review, document and deployment evidence. This task is not complete when the board is populated, the prototype works or tests only cover schemas.

Acceptance:

  • Every leaf and reused requirement has accepted evidence, all critical/high findings are resolved, and the exact candidate is on origin/main with a passing normal Dev release.
  • Approved feature enablement, reversible rollout/rollback, measured performance and operational recovery are demonstrated; any UAT promotion remains Elikem-authorized and is recorded separately.
  • The final report distinguishes implemented, independently verified, deployed and enabled status; no required item is hidden as later work.

Dependencies: eve-interactive-artifacts:M4, eve-interactive-artifacts:QA.01, eve-interactive-artifacts:QA.02, eve-interactive-artifacts:QA.03, eve-interactive-artifacts:DOC.01, platform-ops:eve-artifact-rollout-readiness-2026-10-09.

Code starting points#

Existing entry points include libs/oshun/shell-assistant/src/generative-ui.ts and turn-stream-frames.ts; apps/oshun/bff/src/assistant and routes/assistant.ts; admin/web assistant components; apps/oshun/eve-content/src/service.ts and workspace/PresentationEditor.tsx. New modules must follow repository project/library requirements. Paths are starting points; verify current code and task contracts before edits. When a path is stale, document and resolve the real seam rather than implementing a dummy substitute.